Stampede Theory: Human Nature, Technology, and Runaway Social Realities

By Philip Feldman

Stampede Theory: Human Nature, Technology, and Runaway Social Realities explores the biological, evolutionary and technological systems that drive troubling patterns of behavior among groups while also proposing actions to combat harm. The book discusses different ways that living beings coordinate and how the emergence of communication technologies has changed behaviors. As the problem of echo chambers and misinformation grows, it is crucial to understand underlying causes and provide solutions—this book does just that by pulling from multiple fields to produce a coherent story about how social realities are created and how they can create resilient communities or reinforce damaging beliefs.

This interdisciplinary approach rests on three primary pillars: 1) How information systems affect the distribution of ideas, information, influence and belief; 2. Technology-mediated communication between individuals and groups, from stories pressed into clay tablets to “likes on social media; 3) The sociology of behavioral bias in groups ranging from teams to nations. Because of its interdisciplinary foundations, the book includes chapters that address behavioral economics, cults, artificial intelligence, and the individual psychology of belief. This will be a valuable resource for a range of readers, from political and social scientists to decision-makers in government and business, scientists in the fields of machine learning and AI, and more.

• Presents a usable framework to approach and understand current sociotechnical trends, as well as methods for repair in order to create resilient, sustainable online environments
• Establishes an understanding of the deep biases in human nature and how they interact with technology
• Details specific mechanisms to approach the repair of our sociotechnical systems, including diversity injection and belief cartography

• Language: English
• Publisher: Elsevier Science
• Release date: Apr 26, 2023
• ISBN: 9780443137365

Philip Feldman

Dr. Philip Feldman has spent most of his career building technology for people to use: graphical interfaces, robots, even exercise machines that play video games. He has degrees in art and ecology, as well as a PhD in Human Centered Computing. He has 12 patents, and his academic research focuses on how technology affects why people believe things and how populations make decisions. He has developed several techniques for polling populations using large transformer language models, which allow the latent beliefs of a group to be explored. Dr. Feldman has been a developer and entrepreneur, helping to start small companies that range from medical to virtual reality. He is currently a research professor in Information Systems at the University of Maryland Baltimore County where he studies how to build diverse and resilient systems and that can improve the way we communicate with each other through our ever-present devices.

Book preview

Preface

Philip Feldman     

The summer of 2015 is when I started to fear the internet.

The promise of the Arab Spring had collapsed into the Arab Winter [1]. Something called #GamerGate was harassing women videogame developers and journalists [2]. Trolling and flaming, which had existed since dial-up bulletin board services seemed to be suddenly more organized. Some threshold had been crossed, where online reactionary forces were finding their voice – against democracy in the Middle East and against progressive causes in the West. I felt like the world was being pulled towards an authoritarian future by forces that we don't understand.

The press seemed unprepared to deal with this. Provably false information, amplified online, would find its way into mainstream media reporting. This reached some kind of watershed in the US presidential election of 2016 with the Clinton email scandal. Hillary Clinton had been accused of improperly using a private email server while she was Secretary of State. This scandal dominated the news cycle from May of 2016 until the election, with nearly 70,000 press mentions. Yet, at the same time, people associated with the Trump campaign were engaged in activities that later produced guilty pleas or convictions for six people. Press coverage of Trump was more in line with his themes of immigration (40,000 mentions) and Muslims (30,000 mentions). Only 5,000 mentions were associated with criminal activity on the part of the Trump campaign during the runup to the election [3,4].

This behavior is often explained away by bothsidesism – where a reporter or news organization gives both sides of a story equal weight, even when one side is demonstrably false [5]. But for me it seemed to be something deeper. Why the broad emergence of reactionary themes and conspiracy theories? I have a background in ecology and artificial life, and these patterns seemed more general than some inappropriate applications of a journalistic practice. It reminded me of how populations of animals respond to changes in their environment.

What we believe as individuals emerges from how we think as groups. And in humans, we have two ways of deep biases that structure group though. We can align with a leader, creating the types of dominance hierarchies that we inherited from our Great Ape ancestors. Or we can come to a group consensus, using the more egalitarian, reverse dominance behaviors that we developed as Paleolithic hunter-gatherers. Technology has collided with these two ancient, opposing human traits with the result that for country-sized groups, that balance shifts back and forth, from despots to democracies and back again.

My research on information systems and human-centered computing looks at how technology changes the way human groups interact. Every development in communication technology, from stories told around the campfire to YouTube videos promoting conspiracy theories about vaccines, affects the way we behave around information and each other. This book shows how some changes in technology can support conspiracy theories, while others can produce healthy information behaviors and disrupt belief stampedes around dangerous misinformation.

Stampede populations are at risk for developing dangerous behaviors such as social inertia, where the collective is unable to adapt to a changing environment. In the extreme, this can lead to runaway conditions where a population is so influenced by the social reality that significant numbers of the population can perish [6]. After all, in a stampede, the safest immediate action is often to run with the crowd or risk being trampled, even if there is danger ahead.

Stampede Theory makes the case for a different approach in thinking about misinformation and polarization. We cannot solve behavior problems with fact-checking targeted at individuals. We need to understand how technology interacts with basic human nature like social dominance and egalitarianism. We need to understand how stories spread beliefs through populations. We need to see how our online interactions with people and algorithms affect how we place ourselves in belief spaces, which we construct together with other people and their opinions. Using the data stored in massive Transformer Language models, we can map these relationships, and start to understand and predict locations and movements in these spaces.

This book shows how the simple injection of trustworthy diversity into our search results and social feeds can disrupt belief stampedes. It also describes how to use neural networks and the oceans of online text that we generate every day to construct maps of belief space. Using these maps, we visualize the belief spaces associated with conspiracy theories. We then explore how to track the movement of beliefs across these spaces, and how this can be used to nudge people into more accurate views of their world.

My journey to understand these problems and provide potential solutions led to the book you are holding in your hands. More than anything, that trip is about connecting with people. Starting with my PhD committee: Wayne Lutters, who guided me through this process, from my first review (A wall of ones) through best posters and papers to the point where I'm writing this. I have valued your company on every step of this journey. Don Engel, who brought a physicist's rigor to the simulation work, and unstoppable enthusiasm. Aaron Massey, without whom I wouldn't have had a clue how to address the ethical implications of this work, and for his strong -fu. Shimei Pan for the conversations, encouragement, and for welcoming me into her weekly machine learning group. Lastly, Thom Lieb, who served as my introduction and guide to the journalism profession.

A huge thanks to Aaron Dant – my co-worker, dungeon master, occasional boss, and fellow traveler through this research. I'm not even sure that this book would exist without his unbelievable amount of support. It certainly would have been less fun!

All the friends and acquaintances who helped me think these things through. In particular, Carol Kramme who made me realize that fashion was social thinking. Barbara Schell who got me through my proposal. Stacey Peterson, who added her perspectives as a professor of communications and contributed her observations. My two freelance editors, Michelle Cassidy and Brenda Baer who helped work my ramblings into writing. And, of course, my cycling buddies, who have let me rattle on and on about this work during rides for years now: David Kelling, Roger Eastman, Terry Harrigan, Stuart Lamb, Ricardo Gonzales, Josh Meyers, Ross Chaison, Travis Warren and everyone else who wound up in an impromptu workshop on misinformation at the 50-mile snack break

ASRC Federal, who let me negotiate 20% of my hours to use on research. I literally could not have gotten this far without that charge code to use for research, conferences and writing.

So many authors – Hannah Arendt, Stuart Kauffman, Colin Martindale, Michael Bacharach, Robert Axelrod, Craig Reynolds, Joanna Bryson, Jane Goodall. And that's just the tip of the iceberg. I'm sure that Azimov's Foundation Trilogy and his idea of psychohistory was lurking in my subconscious as these concepts coalesced.

Google Scholar – how did research get done before this? Anurag Acharya [7], if we ever meet the tab's on me.

Suzanne and Vicki for being family. For taking the load when I couldn't be there, and for all the encouragement. And pie.

Alfred and Frances, my parents. Supported and encouraged doesn't begin to describe it. Sadly, they were only able to share in the beginning of this arc, and passed before I finished.

Lastly, my team at Elsevier! Stephen Merken who first read the proposal and passed it on to Katy Eryilmaz who got it approved. Teddy Lewis, who managed the project, Christian Bilbow for the design, Mohan Raj Rajendran, who made sure that I didn't violate the arcania of copyrights, and UnniKannan Ramu, who made sure I got paid!

And the pets – Skip, Andy, and Bennie. "Why are you looking at that when you could be paying attention to meeeeeeeee!"

Baltimore, Maryland

December 2022

References

[1] L. Grinin, A. Korotayev, A. Tausch, Introduction. Why Arab spring became Arab winter, Islamism, Arab Spring, and the Future of Democracy. Springer; 2019:1–24.

[2] T.E. Mortensen, Anger, fear, and games: The long event of #GamerGate, Games and Culture 2018;13(8):787–806.

[3] R. Faris, H. Roberts, B. Etling, N. Bourassa, E. Zuckerman, Y. Benkler, Partisanship, propaganda, and disinformation: Online media and the 2016 US presidential election, Berkman Klein Center Research Publications 2017;6.

[4] R. Mueller, Report on the investigation into Russian interference in the 2016 presidential election. US Department of Justice; 2019.

[5] Bothsidesing: Not all sides are equal, https://www.merriam-webster.com/words-at-play/bothsidesing-bothsidesism-new-words-were-watching.

[6] C.J. Torney, T. Lorenzi, I.D. Couzin, S.A. Levin, Social information use and the evolution of unresponsiveness in collective systems, Journal of the Royal Society Interface 2015;12(103), 20140893.

[7] S. Levy, The gentleman who made scholar, Wired (Jun 2017) https://www.wired.com/2014/10/the-gentleman-who-made-scholar/.

Chapter One: Introduction

Abstract

When we talk about beliefs – political, religious, or just about anything else we have an opinion on, we find it natural to use terms that are deeply tied to location. We can take a position. We have a point of view. We can share common ground or take a leap of faith. Why would this be? Opinions and beliefs are simply thoughts, patterns of electricity and chemistry in our brains. There is nothing about an opinion that has anything to do with location or position.

Moving in physical ways through belief feels natural because we are built that way. We apply instincts developed for the physical world. Changes in belief have an optimal velocity. Too slow and we get bored. Too fast and we become exhausted. Beliefs that change at the right rate make it easier to move together like birds in a flock of fish in a school. A group of people having similar beliefs, oriented towards the same goals are moving together through a shared belief space.

Keywords

Misinformation; Opinion dynamics; Conspiracy theories; Belief space

This is a book about how and why groups of people believe – really, really, believe – in things that do not exist. Things like COVID-19 vaccines have mind-control chips in them, that immigrants are pouring over unprotected borders, that Vladimir Putin has sexually explicit ‘kompromat’¹ on Donald Trump, or that global warming is a hoax. We are living in a time when the dangers of such beliefs should seem painfully obvious, but instead, we are seeing people increasingly drawn to them even when they conflict with their own self-interest.

False beliefs resulted in the Rwandan Genocide of 1994, in which over one million people were slaughtered over the course of 100 days. The 2003 Iraq war was justified by the false belief that Saddam Hussein was hiding weapons of mass destruction. During the COVID-19 epidemic, the false belief that the vaccines were somehow more dangerous than the disease was shared by millions of people. Thousands upon thousands of these people died, slowly suffocating as their lungs disintegrated into red pulp. Many succumbed never admitting that they had been wrong. Why?

To understand how such large groups can organize themselves around things that could clearly be shown to be false at the time, you need to understand that belief is a place.

When we talk about beliefs – political, religious, or just about anything else we have an opinion on, we find it natural to use terms that are deeply tied to location. We can take a position. We have a point of view. We can share common ground or take a leap of faith. Why would this be? Opinions and beliefs are simply thoughts, patterns of electricity and chemistry in our brains. There is nothing about an opinion that has anything to do with location or position.

And yet, language and literature are full of this fundamental tendency on the part of humans to locate themselves within their beliefs. This process is not limited to how we feel about information. The way we create our works of fiction also seem to rely on this same sense of location and position. Stories are journeys, which have plot lines and arcs.

Information, the thing that underlies belief, has no inherent place. It is simply a collection of data. Computers, unlike us, do not feel obligated to organize their outputs into narratives. But when humans talk about data and science, we tell stories that place these discoveries in a landscape of ideas. The way that we as groups understand information is inherently narrative. We tell and retell stories to each other. How those stories evolve and change, a little here, a lot there, is an example of how we collectively build belief structures in which we dwell, move within, and reshape from time to time.

The use of location and movement in our stories seems to be a natural consequence of the fact that human beings are fundamentally location-based animals. For virtually all of the time that life has been on Earth, the only reality that mattered was physical. Humans are the first organisms whose lives depend as much on abstract information as the physical environment.

It is unlikely that we could develop any other system, given the brief time in our evolution that there has even been language, much less the concept of information [8]. As our brains evolved, they adapted the structures that they already had to new, more abstract purposes. We are the decedents of a long line of organisms that are deeply tied to a physical environment, and this is reflected in our language, our stories, and how we understand information.

Our beliefs are places in a shared terrain.

We can have locations in that terrain. We can move in that landscape. Groups can form and break up. In belief space, just like physical space, we can be nomads, moving to the beat of our own drummer, like Roald Amundsen, the Norwegian explorer who beat the much better funded British expedition of Robert Scott to the South Pole [9]. We can flock with others and create fashions in everything from clothing to programming languages. We can lose ourselves in a stampede, careening mindlessly with others away from danger, like during the 1913 Italian Hall disaster when a false shout of Fire led to a panic resulting in 73 deaths, including women and children² [10].

In the same way that technology such as maps and GPS have profoundly affected the way we move in the world (can you really be lost anymore?), technology has affected the way we behave in this space too (can you really be offline anymore?).

Moving in physical ways through belief feels natural because we're built that way. We apply instincts developed for the physical world to virtual group coordination [11]. Changes in belief have an optimal velocity. Too slow and we get bored – we like novelty. Too fast and we become exhausted [12] – we don't like getting overwhelmed. Beliefs that change at the right pace make it easier to move together [13], like birds in a flock or fish in a school. A bunch of people having similar beliefs, oriented towards the same goals are moving together through belief space. It's not surprising that they are going to find it easier to do things as a group [14]. You may find it surprising that our brains synchronize, actually firing at the same times and the same places when we are sharing a story. And you can tell by looking at brain scans that people are exploring, flocking, or stampeding together in these virtual spaces [13].

These patterns of motion are greatly influenced by the communications media through which we coordinate. Small villages connected only by foot traffic behave differently from globally interconnected populations of humans, synthetic agents, and information retrieval systems.

In addition, human groups behave on a spectrum from communal to hierarchical. Hierarchical structures produce stability and order. Egalitarian communities of peers are reverse hierarchies that prevent despotic leaders from emerging [15]. Egalitarian communities better support creativity and diversity. The way these natural systems interact with technology have produced complex, emergent behavior, starting with writing, continuing with mass communication, and our current web-based social networks.

Through most of our history, early and modern humans have lived in groups that ranged in size from tens to hundreds, and communication has been physical and direct. Communal and hierarchical structures existed in shifting balance. As we'll see later in Section 3.4, technology influenced this balance, starting with the development of our most primal technologies language, weapons, and fire. For the first time, a group could find out what had happened somewhere else through language, and the members of that group could confront the offender with weapons. These developments shifted the balance towards egalitarianism. The later development of agriculture and the domestication of animals shifted the balance back to hierarchies. Technology continues to affect how we behave as groups. It can increase the capacity for complex ideas to emerge from groups of peers or increase the hierarchical domination of one group over others.

Individuals incorporate the massive complexity of physical and social reality into personal narratives. Groups integrate these into common knowledge, which leads to common behaviors. This is what I call thinking as populations. The process of simplifying the complexity in the world to a representation that supports effective coordination is very general and has deep implications ranging from human and animal group behavior to artificial intelligence. We'll look at a wide variety of examples, from animals on the Serengeti, to how money developed, to the surprising initial story of the Tower of Babel and how it applies to a spacecraft disintegrating in the Martian atmosphere.

We will explore how the process of building consensus can produce social facts that are often more powerful than any objective reality. We will also examine the interplay of independent thinkers and group players, of fashion leaders and followers, and how our modern information technologies are profoundly influencing our ability to navigate across these spaces.

We're going to look at the pieces that make up this mixture of instinct and technology. We'll start with the basic rules of how individuals and groups behave in the presence of expensive information, and how that leads to beliefs ranging from science to religion to cults. We'll then look at how communication technology interacts with our fundamental biases as tool-using primates, and how that can create population-scale information pathologies.

Lastly, we'll consider how these fundamental rules affect the development of Artificial Intelligence. The machines that we have built to manipulate information do not share our physics-based biases about how to interpret the world. The way that they store this information is fundamentally new. And the relationships between all the stories, posts, books, and articles are fixed in billions of values etched in computer memory. These machines can create stories based on what they have learned. And we can use these stories as a new way to understand ourselves.

We can't ask people to carefully describe their beliefs in repeatable ways. People get bored. They often tell researchers what they want to hear. They can change their minds. But in these giant AI models, those billions of values representing stories, posts, books, and articles are fixed. We can explore these relationships in ways that we never could before. Given a single starting point, these machines can generate countless narrative trajectories through these spaces. We can knit these trajectories into maps that reveal these relationships.

Later in this book, we'll look at some preliminary work in this area that builds maps of the places where conspiracy theories live. We'll look at some of the most popular conspiracies, and we'll follow the stories that connect them. We'll see how these beliefs relate to each other, and we'll learn that mapping these spaces is a way to better understand what we believe, who we are, and how we fit into the world around us. This new way of looking at ourselves, and it opens up ways of identifying and disrupting dangerous belief stampedes in ways that are surprisingly simple, and effective.

Belief is a place, and we are on the verge of finally being able to see where we've been living for all these years.

References

[8] I. Davidson, The archaeology of language origins – a review, Antiquity 1991;65(246):39–48.

[9] R. Huntford, The last place on earth: Scott and Amundsen's race to the south pole, Abacus 2000:546–547.

[10] K.M. Ngai, F.M. Burkle, A. Hsu, E.B. Hsu, Human stampedes: A systematic review of historical and peer-reviewed sources, Disaster Medicine and Public Health Preparedness 2009;3(4):191–195.

[11] G.F. Young, L. Scardovi, A. Cavagna, I. Giardina, N.E. Leonard, Starling flock networks manage uncertainty in consensus at low cost, PLoS Computational Biology 2013;9(1), e1002894.

[12] C. Martindale, The Clockwork Muse: The Predictability of Artistic Change. New York: Basic; 1990.

[13] G.J. Stephens, L.J. Silbert, U. Hasson, Speaker–listener neural coupling underlies successful communication, Proceedings of the National Academy of Sciences 2010;107(32):14425–14430.

[14] S. Moscovici, W. Doise, Conflict and Consensus: A General Theory of Collective Decisions. Sage; 1994.

[15] C. Boehm, H.B. Barclay, R.K. Dentan, M.-C. Dupre, J.D. Hill, S. Kent, B.M. Knauft, K.F. Otterbein, S. Rayner, Egalitarian behavior and reverse dominance hierarchy [and comments and reply], Current Anthropology 1993;34(3):227–254.

---

¹  https://en.wikipedia.org/wiki/Kompromat.

²  This incident contributed to the US Supreme Court deciding in 1919 declaring the act falsely shouting fire in a theatre and causing a panic unconstitutional.

Part One: Theory

Outline

Chapter Two. From the Serengeti to the Ecclesia

Chapter Three. Deep bias

Chapter Four. Humans and information

Chapter Five. Human belief spaces

Chapter Six. Influence + dominance = attention

Chapter Seven. Hierarchies, networks, and technology

Chapter Two: From the Serengeti to the Ecclesia

Abstract

The ability to allocate responsibility is a powerful quality of social beings, but how a group operates varies depending on the type of group and their desired outcome. Bees coordinate their large populations by having extremely specialized roles. Herds of wildebeest and schools of sardines use mutual influence. Apes add hierarchies to the mix. This chapter sets the stage for the book by examining populations from the animal world and as well as styles of human organizations, from authoritarian to egalitarian.

Keywords

Group coordination; Hypersocial; Mutual influence; Hierarchy

Collecting good information and using it to make decisions can be expensive. Paying attention to one thing often happens at the expense of another. It can take time to accumulate what is needed to make a decision, time that might have been better spent on other things. For example, a wildebeest on the Serengeti plains has multiple competing problems that need to be solved: Where do I eat? Are there predators nearby? When can I sleep? A mistake here can easily be fatal.

The ability for social groups to share responsibility is a powerful thing. Grazing animals like our wildebeest on the Serengeti are seldom solitary. They will informally distribute sentry duty among members of the herd who look for predators so the rest of the herd can concentrate on other activities. The herd trusts that these sentries can detect threats so they can eat, sleep, and mate. The herd thrives because of this kind of distributed responsibility. Just as some look for threats, others indicate when they've found food. The whole outperforms the sum of its parts.

Spontaneous, yet coordinated group behavior is known as swarm intelligence. Swarms are networks with no leader, and they are capable of extremely sophisticated behavior. For example, a murmuration of starlings decides what to do as a group in response to predators like hawks or deciding where to land and spend the night. Most of this behavior emerges simply from each starling adjusting its flight in response to what it sees of the world and the behavior of its closest neighbors